Also strongly recommend the adjacent possible work if you haven't seen it yet.
One of the main issues I have with economic approaches to knowledge production like Romer's is that they approach Knowledge as a singular entity that produces an aggregate economic output, as opposed to a set of n knowledges that each have their own output curves. In practice (and in the technological forecasting lit), there's a recognition of this fact in the modelling of new technologies as overlapping S-curves:
The S-curve approach recognizes that there is a point where the ROI for continued investment in a particular technology tapers off. So this means that we don't run out of Ideas, we just overinvest in exploitation of existing ideas until the cost-to-ROI ratio becomes ridiculous. Unfortunately, the structure of many of our management approaches and institutions are built on extraction and exploitation of singular ideas, and we are Very Bad at recognizing when it's necessary to pivot from exploit to explore.
Even worse, this is not only true for the structure of our industrial-commercial research, but for our basic research apparatus as well. Grants are provided on the basis of previous positive results, as opposed to exploration of net-new areas -- resulting in overexploitation of existing ideas and incentivized research stagnation.
In practice, you can think of it like this comic...
except instead of a circle, our knowledge distribution looks like this
There are a few issues to parse here --
Prizes and direct subsidies already exist for high-CAPEX projects and are also a crucial part of the system as it exists today. Almost all biopharma research starts out as state-funded work out of universities or NIH, until it reaches a point where it's ready for trials and gets spun out into a business. The trials themselves are then often further subsidized. There's a lot of work and money that has to go into making biotech viable just because of how little we understand about fundamental biology.
The exclusivity rights conferred by intellectual property are exceptionally important in areas where the innovator's CAPEX requirements are high, but the cost to copycats is low -- like biopharma, energy, or even mining. The creation of a temporary monopoly engineers sufficient market value to make initial investment of high-risk capital on the order of hundreds of millions (if not billions) of dollars worthwhile and palatable (e.g. see https://dukespace.lib.duke.edu/dspace/bitstream/handle/10161/6707/Record5.pdf; https://digitalcommons.law.scu.edu/cgi/viewcontent.cgi?article=1019&context=facpubs ).
The rules were never supposed to apply to extremely low CAPEX industries like software, however -- hence why "abstract ideas" were excluded from intellectual property law and are often negotiated and renegotiated in the courts as incumbent monopolies and patent trolls try to use regulatory capture to maintain their status.